Lithography Hotspot Detection Based on Improved Yolov5s

Qingyue Wu; Jiamin Liu; Song Zhang; Hao Jiang; Shiyuan Liu

doi:10.3788/LOP231038

Laser & Optoelectronics Progress, Volume. 60, Issue 24, 2422001(2023)

Lithography Hotspot Detection Based on Improved Yolov5s

Qingyue Wu¹, Jiamin Liu^1、*, Song Zhang¹, Hao Jiang¹, and Shiyuan Liu^1,2、**

¹State Key Laboratory of Intelligent Manufacturing Equipment and Technology, Huazhong University of Science and Technology, Wuhan 430074, Hubei, China

²Hubei Optics Valley Laboratory, Wuhan 430074, Hubei, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Network structure of Yolov5s

Download full size

Fig. 2. CA module structure

Download full size

Fig. 3. C3 module structure before and after introduction of CA mechanism. (a) C3 module structure before introduction of CA; (b) C3 module structure after introduction of CA

Download full size

Fig. 4. Comparison between SiLU function and Leaky ReLU function

Download full size

Fig. 5. Improved CBS module structure

Download full size

Fig. 6. Improved hotspot detection network model structure

Download full size

Fig. 7. Geometric features of different types of lithography hotspots

Download full size

Fig. 8. Distribution information of six types of lithography hotspots

Download full size

Fig. 9. Comparison of detection results of Yolov5s and improved Yolov5s in different lithography hotspot layouts. (a)‒(f) Detection results of layout 1 to 6

Download full size

Table 1. Network training parameters
View table
Table 1. Network training parameters
Parameter Value
Epoch 500
Batch size 16
IoU threshold 0.65
Initial learning rate 0.01
Minimum learning rate 0.002
Momentum 0.937
Weight decay 0.0005

Table 2. Model detection results

View table

Table 2. Model detection results

Type	P /%	R /%	f_F1-score/%	f_mAP/%
Missing	99.7	95.2	97.4	95.7
Extra	100.0	92.8	96.3	97.3
Soft bridging	93.4	100.0	96.0	98.9
Hard bridging	97.9	100.0	98.9	99.5
Soft pinching	97.2	100.0	98.6	99.5
Hard pinching	97.9	100.0	98.6	99.5
Average	97.7	98.0	97.8	98.4

Table 3. Ablation experiment results

View table

Table 3. Ablation experiment results

Method				P /%	R /%	f_F1-score /%	f_mAP /%	Runtime /（h·mm^-2）
Yolov5s	SiLU	SIoU	CA	P /%	R /%	f_F1-score /%	f_mAP /%	Runtime /（h·mm^-2）
√				96.4	95.2	95.8	97.3	0.687
√	√			95.6	97.2	96.4	97.9	0.610
√		√		96.9	96.9	96.9	98.1	0.687
√			√	98.6	96.0	97.3	98.0	0.625
√	√		√	97.5	97.3	97.4	98.2	0.740
√	√	√	√	97.7	98.0	97.8	98.4	0.697

Table 4. Comparison of experimental results of different methods
View table
Table 4. Comparison of experimental results of different methods
Method P /% R /% f_F1-score /%
Runtime /
（h·mm^-2）
CNN 16.0 97.8 26.5 25.2
GoogLeNet 47.5 98.3 63.5 —
Faster R-CNN 40.3 96.1 53.1 0.7
Proposed method 97.7 98.0 97.8 0.7

Tools

Get Citation

Copy Citation Text

Qingyue Wu, Jiamin Liu, Song Zhang, Hao Jiang, Shiyuan Liu. Lithography Hotspot Detection Based on Improved Yolov5s[J]. Laser & Optoelectronics Progress, 2023, 60(24): 2422001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites